Know Nota
Sign inTry it free
For organisations

Know what you are publishing.

Before a piece runs, a hire is made or a claim goes out, Nota tells you how machine-like the text reads — with how often the model is wrong beside the score — and gives you a signed record you can file. Your writers can attach a record of how they wrote it. No verdicts, no accusations, nothing stored unless you keep it.

Try Nota free →How it works
01

A number you can act on, with its error rate

Every result says how often the exact model that scored it is wrong about real human writing. Your policy can name a line — “look closer above elevated, when the error rate is under 2%” — instead of trusting a tool that hides its misses.

02

A record that holds up later

Each scan can produce a signed Nota Record: the reading, the model version, the fingerprint of the exact text, the warnings. Signed by us, checkable by anyone with the link — file it with the piece, the contract or the application.

03

Ask writers to show their work, not to prove innocence

A freelancer or a candidate brings a Writing Record — what the document’s own history says about how it was written, plus their own declaration of what tools they used. It turns a suspicion into a paper trail, on their terms.

What you get to work with.

Batch up to 100 files

Drop a folder of drafts, applications or submissions. Per-document results: words, score, band, how often the model is wrong.

Source check

Where the words also appear on the web and in scholarly sources, with the matched passages highlighted. Evidence of overlap, never a plagiarism verdict.

API with the warnings built in

Bearer keys, usage by endpoint and day, the same JSON as the app — score, band, how often it is wrong, model version, warning. Metered by Nota words.

Nothing stored unless you keep it

Text is processed in memory and discarded. Records keep the fingerprint and the numbers; the text only if you choose to keep it in the record.

On every result, whatever the plan: probability, band, threshold, measured false-positive rate, model version, word count, caveat. Never “written by AI”.

Fair questions

The longer answers live on the method page.

Can we reject a piece or a candidate on a Nota score?We would not. A score is a reading of the text with a known error rate — a reason to look closer, not a decision. Pair it with the source check, the Writing Record, and a conversation. The warning saying so is printed on every result and inside every record.
Does Nota keep our documents?No. Text is processed in memory and discarded. A signed record keeps the fingerprint and the numbers; if you choose to keep the text inside the record, it lives in your account, not in a training set. Nothing is ever used for training.
What does the API return?The same JSON the app shows: score, band, how often the model is wrong at this length, the model version, the word count, and the warning. Bearer keys on the Professional plan; prepaid credits for volume.
Can our writers use it too?Yes — the Writing Record is free on every plan. A writer brings the .docx (or, soon, the Google Doc), Nota reads how it was written, and they get a signed link to attach. Honest AI use can be declared on the record instead of hidden.
Do you sell an accuracy number?We publish how often the model is wrong, measured on real human writing of 250 words or more, and it comes back with every result. We do not print a single flattering “accuracy” figure — those hide the misses.

Run a batch today.

Individual and Professional plans meter Nota words; the API is on Professional. Teams with seats and shared quota come later — talk to us.

See plans →